NUMA memory latency is much, much less of a big deal than it used to be back in the early days of NUMA box vendors like Sequent. Back in the bad old days remote memory was something like 50x slower than local.
Now it's much faster - the default memory configuration on most two way NUMA boxes is to interleave pages across both CPUs. This is slower on average than a NUMA aware application but not slow enough to cause a significant issue in most cases.